Open andykaufseo opened 6 months ago
Eager to see this issue resolved.
@Quidam2k @andykaufseo I was able to fix this issue. I'm submitting a PR but if you want the fix now, you can copy the commit from my fork. It's very simple. https://github.com/halkony/sd-forge-layerdiffuse
The root cause is the way the image is handled inside the txt2img_function
(which is run whenever you generate images). Instead of storing the processed image in PIL format, it's stored as a numpy array in processed.extra_images
. My PR just adds a hook after processing that converts it and saves it with the tag "transparent."
Looks like there's another PR in the pipeline that fixes this issue and has some other fixes.
https://github.com/layerdiffusion/sd-forge-layerdiffuse/pull/99
Author seems inactive -- so until that's merged, using this fork is preferable https://github.com/new-sankaku/sd-forge-layerdiffuse/tree/main
The extension works fine, it generates transparent images, but the only way to save the transparent ones is to right click > save on each image, which is not a decent workflow for generating bulk images.
The script actually saves the images without the transparent bg to the output folder, it saves the ones with the checkered background.